An artificial intelligence algorithm for the classification of sphenoid sinus pneumatisation on sinus computed...

Area of Science:

• Radiology
• Artificial Intelligence
• Otolaryngology

Background:

• Sphenoid pneumatization classification is crucial for sinus CT reporting.
• Artificial intelligence (AI) and convolutional neural networks (CNNs) show promise in radiological image classification.

Purpose of the Study:

• To evaluate the accuracy of a trained CNN in classifying sphenoid sinus pneumatization on CT imaging.

Main Methods:

• Retrospective analysis of sagittal CT slices of the sphenoid sinus.
• Classification of pneumatization into conchal, presellar, or sellar types by blinded reviewers and an AI algorithm.
• Development of the AI algorithm using Microsoft Azure Custom Vision.

Main Results:

• The CNN achieved high accuracy in classifying sphenoid pneumatization subtypes: 93.2% for conchal, 87.3% for presellar, and 85.6% for sellar.
• Overall weighted accuracy of the CNN was 85.9% on test data.
• The algorithm was trained on 780 images and tested on 118 images.

Conclusions:

• The developed CNN demonstrates moderate accuracy in classifying sphenoid pneumatization subtypes.
• CNN-based assistive tools can enhance surgical planning through automated reporting.
• Automated classification allows for increased focus on identifying sinonasal pathology.